I’m still very busy with my software work, but since urban crime and violence—often with a sharp racial tinge—are regularly in the headlines, I’m republishing my major Race/Crime article from a few years ago.
Among other things, I presented what might possibly be the largest empirical correlation discovered anywhere in the modern social sciences.
The noted science fiction writer Philip K. Dick once declared that “Reality is what continues to exist whether you believe in it or not.” Such an observation should be kept in mind when we consider some of the touchier aspects of American society.
Recall the notorious case of Daniel Patrick Moynihan, whose 1965 report on the terrible deterioration in the condition of the black American family aroused such a firestorm of denunciation and outrage in liberal circles that the topic was rendered totally radioactive for the better part of a generation. Eventually the continuing deterioration reached such massive proportions that the subject was taken up again by prominent liberals in the 1980s, who then declared Moynihan a prophetic voice, unjustly condemned.
This contentious history of racially-charged social analysis was certainly in the back of my mind when I began my quantitative research into Hispanic crime rates in late 2009. One traditional difficulty in producing such estimates had been the problematical nature of the data. Although the FBI Uniform Crime Reports readily show the annual totals of black and Asian criminal perpetrators, Hispanics are generally grouped together with whites and no separate figures are provided, thereby allowing all sorts of extreme speculation by those so inclined.
In order to distinguish reality from vivid imagination, a major section of my analysis focused on the data from America’s larger cities, exploring the correlations between their FBI-reported crime rates and their Census-reported ethnic proportions. If urban crime rates had little relation to the relative size of the local Hispanic population, this would indicate that Hispanics did not have unusually high rates of criminality. Furthermore, densely populated urban centers have almost always had far more crime than rural areas or suburbs, so restricting the analysis to cities would reduce the impact of that extraneous variable, which might otherwise artificially inflate the national crime statistics for a heavily urbanized population group such as Hispanics.
My expectations proved entirely correct, and the correlations between Hispanic percentages and local crime rates were usually quite close to the same figures for whites, strongly supporting my hypothesis that the two groups had fairly similar rates of urban criminality despite their huge differences in socio-economic status. But that same simple calculation yielded a remarkably strong correlation between black numbers and crime, fully confirming the implications of the FBI racial data on perpetrators.
This presented me with an obvious quandary. The topic of my article was “Hispanic crime” and my research findings were original and potentially an important addition to the public policy debate. Yet the black crime figures in my charts and graphs were so striking that I realized they might easily overshadow my other results, becoming the focus of an explosive debate that would inevitably deflect attention away from my central conclusion. Therefore, I chose to excise the black results, perhaps improperly elevating political prudence over intellectual candor.
I further justified this decision by noting that black crime in America had been an important topic of public discussion for at least the last half-century. I reasoned that my findings must surely have been quietly known for decades to most social scientists in the relevant fields, and hence would add little to existing knowledge. However, since that time a few private discussions have led me to seriously question that assumption, as has the emotion-laden but vacuous media firestorm surrounding the George Zimmerman trial. I have therefore now decided to publish an expanded and unexpurgated version of my analysis, which I believe may have important explanatory value as well as some interesting policy implications.
My central methodology is simple. I obtained the crime rates and ethnic percentages of America’s larger cities from official government data sources and calculated the population-weighted cross-correlations. In order to minimize the impact of statistical outliers, I applied this same approach to hundreds of different datasets: each of the years 1985 through 2011; homicide rates, robbery rates, and violent crime overall; all large cities of 250,000 and above and also restricted only to major cities of at least 500,000. I obtained these urban crime correlations with respect to the percentages of local whites, blacks, and Hispanics, but excluded Asians since their numbers were quite insignificant until recently (here and throughout this article, “white” shall refer to non-Hispanic whites).
I also attempted to estimate these same results for the overall immigrant population. The overwhelming majority of immigrants since 1965 have been Hispanic or Asian while conversely the overwhelming majority of those two population groups have a relatively recent immigrant family background. So the combined population of Hispanics and Asians constitutes a good proxy for the immigrant community, and allows us to determine the immigrant relationship to crime rates.
Presented graphically, these various urban crime correlations are as follows:
These charts demonstrate that over the last twenty-five years the weighted correlations for each of the crime categories against the percentages of whites, Hispanics, and “immigrants” (i.e. Hispanics-plus-Asians) have fluctuated in the general range of -0.20 to -0.60. Interestingly enough, for most of the last decade the presence of Hispanics and immigrants has become noticeably less associated with crime than the presence of whites, although that latter category obviously exhibits large regional heterogeneity. Meanwhile, in the case of blacks, the weighted crime correlations have steadily risen from 0.60 to around 0.80 or above, almost always now falling within between 0.75 and 0.85.
These particular calculations do rely upon several minor methodological choices. For example, I have used the 2000 Census population thresholds for selecting the sixty-odd large cities in my dataset, while I could have chosen some other year instead. The substantial annual fluctuations in the urban ethnic percentages provided by the Census-ACS estimates led me to instead use the interpolated Census figures for all years. The annual urban population totals used by the FBI sometimes differ slightly from the Census numbers, and I used the former for population-weighting purposes. However, all my results were quite robust with respect to these particular decisions, and modifying them would produce results largely indistinguishable from those presented above.
On a more difficult matter, there is always the possibility of local bias in FBI crime statistics, with the data for some cities possibly being more reliable or comprehensive than for others. But the reporting rate for homicides is widely accepted as close to 100 percent, and the close correspondence between the results for this “gold standard” crime category and those for the robbery and violent crime rates tends to confirm the validity of the latter. In any event, we would expect the highest-crime areas to be those most likely to suffer from under-reporting problems, so we would expect our figures to somewhat underestimate the true size of the correlations.
It is important to recognize that within the world of academic sociology discovering an important correlation in the range of 0.80 or above is quite remarkable, almost extraordinary. And even these correlations between black population prevalence and urban crime rates may actually tend to significantly understate the reality. All these correlations were performed on a city-wide aggregate basis. The New York City numbers include both the Upper East Side and Brownsville, Los Angeles both Bel Air and Watts, Chicago the Gold Coast and Englewood, with each city’s totals averaging those of both the wealthiest and the most dangerous districts. This crude methodology tends to obscure the local pattern of crime, which usually varies tremendously between different areas, often roughly corresponding to the lines of racial segregation. It is hardly a secret that impoverished black areas do have far higher crime rates than affluent white ones.
If instead we relied upon smaller geographical units such as neighborhoods, our results would be much more precise, but ethnicity data is provided by zip code while crime data is reported by precinct, so a major research undertaking would be required to match these dissimilar aggregational units for calculation purposes. However, the apparent geographical pattern of crime in these cities and most others might lead us to suspect that our national racial correlations would become substantially greater under such a more accurate approach, perhaps often reaching or even exceeding the 0.90 level. The inescapable conclusion is that local urban crime rates in America seem to be almost entirely explained by the local racial distribution.
But could such a strikingly simple sociological truth possibly be correct? After all, academic scholars have long advanced a wide variety of different socio-economic explanations for crime, and these have often been heavily promoted by pundits and the media. Commonly cited factors have been urban density, especially in the case of high-rise housing projects, and local poverty. There is also the relative number of police officers to consider. We should certainly compare the possible influence of these factors with the ethnic ones examined above.
Since the geographical borders of a city are generally fixed, average population densities are easy to calculate and in recent years their apparent impact upon crime rates has been negligible, whether for homicide, robbery, or violent crime in general. For the last dozen years, the density/crime correlations have always ranged between 0.20 and -0.20 and were usually close to zero. Perhaps many of us have an intuitive mental image of densely populated East Coast cities being natural hotbeds of crime. But this appears incorrect: crime rates and urban density seem to have little connection.
What about the sizes of the various urban police departments? Although precise comparisons are sometimes difficult, the Bureau of Justice Statistics periodically publishes official reports on the subject, and the latest 2007 study lists the numerical totals of America’s fifty largest urban police forces, allowing us to calculate the weighted correlations between these per capita policing levels and the corresponding crime rates of the years 2007-2011. We discover that there actually exists a moderately strong positive correlation, generally falling in the range 0.30-0.60: the more police, the more crime. Although this might seem counterintuitive, the explanation becomes obvious once we reverse the direction of causation. Higher crime rates usually persuade local authorities to hire additional police officers.
Finally, although urban crime rates do track local economic conditions, the relationship is far from tight. For the years 2006-2011, the Census-ACS provides estimates of the Mean Income, Median Income, and Poverty Rates for each urban center, and we can easily perform the same calculations we did in the racial case. The correlations between the Mean Income and Median Income levels and the various crime categories generally fall in the range of -0.40 to -0.60, being moderately rather than strongly negative. Even the correlation between Poverty Rate and crime—supported by the obvious truism that most street criminals are poor—is hardly enormous, falling between 0.50 and 0.70, and usually well below our racial figures.
The relative strength of these different correlations may be seen by a chart superimposing the economic and ethnic results for the last dozen years of robbery rate correlations for our major cities. Although the hard economic times since 2008 have considerably increased the influence of the poverty correlate, that factor is still considerably less significant than the racial one.
Indeed, the race/crime correlation so substantially exceeds the poverty/crime relationship that much of the latter may simply be a statistical artifact due to most urban blacks being poor. Consider that both blacks and Hispanics currently have similar national poverty rates in the one-third range, more than double the white figure, and each constitutes well over 20% of our urban population. However, major cities with substantial poverty but few blacks usually tend to have far lower levels of crime. For example, El Paso and Atlanta are comparable in size and have similar poverty rates, but the latter has eight times the robbery rate and over ten times the homicide rate. Within California, Oakland approximately matches Santa Ana in size and poverty, but has several times the rate of crime. Thus, it seems plausible that removing the black population from our calculation might actually reduce the residual poverty/crime correlation for non-blacks to a moderate or even a low figure.
To some extent, this surprising possibility is merely a statistical syllogism. Whenever the correlation to a single factor approaches unity, no other non-equivalent item may have a large, independent impact. And failing to recognize the existence of such a single, overwhelming factor might lead us to misidentify numerous other spurious influences, whose apparent causal importance actually derives from their own correlations with the primary item. For many years, the black connection to local crime has been so strong as to almost eliminate the possible role of any other variable.
We must obviously be cautious in interpreting the meaning of these statistical findings since correlation does not necessarily imply causation. Over the last few years the crime correlation for Hispanic or Hispanic-plus-Asian numbers has been substantially more negative than the same figure for whites, but this does not necessarily prove that whites are much more likely to commit urban crime, though it would tend to rule out the contrary possibility that Hispanics or immigrants have far higher rates of criminality.
However, if we examine the official FBI arrest statistics, we find that these seem to support the most straightforward interpretation of our racial crime correlations. For example, blacks in America were over six times as likely to be arrested for homicide in 2011 as non-blacks and over eight times as likely to be arrested for robbery; the factors for previous years were usually in a similar range. The accuracy of this racial pattern of arrests is generally confirmed by the corresponding racial pattern of victim-identification statements, also aggregated by the FBI. Indeed, several years ago the liberal Sentencing Project organization estimated that some one-third of all American black men are already convicted criminals by their 20s, and the fraction would surely be far higher for those living in urban areas.
A sense of the real world impact of these grim statistics may be found in the stratified 2011 Census-ACS data for major American cities. The three urban centers with the largest black populations are New York City, Chicago, and Philadelphia, and together they contain over one-third more adult black women than black men. The corresponding national shortfall of black males runs well into the millions, partly accounting for the notorious “marriage gap” problems faced by women of their background. Those millions of missing black men are generally dead or in prison.
Over the last few years, the official publications of the Bureau of Justice Statistics have made it increasingly difficult to determine the racial totals of inmates in state prisons and local jails but the figures from the mid-2000s probably still provide a reasonable estimate, and I had used these in my 2010 article. Since crime is overwhelmingly committed by young males, for comparative purposes we should normalize all these incarceration totals against the base population of adult males in their prime-crime years, and the results are summarized in my previously published chart, reprinted here.
Since the mid-1990s, the issue of street crime has mostly dropped off the front pages of our national newspapers and disappeared from the public debate. Meanwhile, black Americans have gained much greater visibility in the upper reaches of our national elites, while Barack Obama has been elected and reelected as our first black president. This might seem to indicate that traditional racial cleavages in our society have become less substantial. Furthermore, with such enormous numbers of young black men now in prison, we might naturally expect that the racial character of American urban crime rates has sharply declined over the last couple of decades. However, the quantitative evidence demonstrates the exact opposite situation, as may be seen by examining the combined twenty-five year trajectories of our various racial crime correlations, which have steadily grown more extreme. The images shown on our film screens or television sets may portray one America, but the actual data reveals a very different country.
Once we accept the reality of these stark racial facts, we must naturally wonder about the causes, and also why the historical trends seem to have been moving in exactly the wrong direction over most of the last quarter-century. Certainly many theoretical explanations have been advanced, both from the Left and the Right, and whole library shelves have been filled with books on the subject since the urban violence of the 1960s. A short article is no place for me to summarize such a vast literature on a contentious topic, especially when I can provide no original insights of my own. But good theoretical analysis requires a solid factual grounding, and my main purpose here is to establish those facts, which others may then choose to interpret howsoever they wish. Absent such information, any national dialogue becomes an exercise in empty ideological posturing.
Racial issues have traditionally been among the most highly charged in American public life, and the nexus of crime and race has been exceptionally contentious for many decades. Under these circumstances respectable scholars tend to be cautious in discussing or merely investigating this topic, and the mainstream media is usually even more gun-shy. The striking racial findings presented above require only trivial statistical calculations and may be glimpsed in any casual inspection of the crime rankings of our major cities. But I remain uncertain to what extent they are already recognized by our experts in social policy.
For example, when I presented my correlation results to one very prominent conservative social scientist, he found them shocking and remarkable, and said he had never imagined that the statistical relationship between race and crime was so extremely strong. But when I showed the same data to an equally prominent liberal academic, he took the information in stride and said he assumed that almost all experts were already quietly aware of the general facts. The reactions of other knowledgeable individuals fell all along this spectrum ranging from surprise to familiarity. Knowledge so explosive that it is usually unspoken and unreported may easily remain unknown even to many of our foremost intellectuals.
But whether or not most of our ruling elites explicitly recognize the stark racial character of American crime, the reality still exists, and we should consider exploring whether these unpublicized facts may have had broader influences in our society, possibly in seemingly unrelated areas. After all, urban crime has frequently been a leading issue in American public life, during some periods ranking as one of the most important. Certain matters may not be easily discussed in polite company these days, but if even just a portion of the citizenry is intuitively aware of the situation, their attitudes might have broader ripple effects throughout the entire population. Is there any substantial evidence for this?
Consider the electoral behavior of American whites, and especially their inclination to support either Democratic or Republican candidates. Because of gerrymandering, most individual congressional districts are overwhelmingly aligned with one party or another, and general elections are a mere formality; this is often also true of statewide races for senator or governor. However, in presidential elections both parties almost always field viable national candidates with a reasonable chance of winning, so these provide the best means of gauging white political alignment. And for these campaigns, the racial lines are clearly established, with the modern Republicans being the “white party,” drawing over 90% of their support from that demographic group, while over 90% of blacks regularly vote the Democratic ticket, which also usually attracts the overwhelming majority of other non-white voters.
As I pointed out in a 2011 article, there has been a striking statewide pattern to white voting behavior over the last couple of decades. Many conservative activists and media pundits have spent years attacking immigrants, illegal or otherwise, and have regularly denounced the cultural threat posed by the growing population of non-English-speakers or non-white foreigners. Nevertheless, the empirical fact is that presence or absence of large numbers of Hispanics or Asians in a given state seems to have virtually no impact upon white voting patterns. Meanwhile, there exists a strong relationship between the size of a state’s black population and the likelihood that local whites will favor the Republicans. The weighted-average correlations between the racial compositions of the fifty states and the degree to which their white voters favor Republican presidential candidates is summarized in the following chart.
GOP leaders are always fearful of being denounced as “racist” by the major media, and often seek to camouflage the underlying source of their electoral support by adopting the most extreme forms of tokenism, promoting black party leaders and spokesmen while heavily recruiting black candidates and focusing almost entirely upon non-racial issues. Conservative activists often rhetorically identify themselves as heirs to the “party of Lincoln” and may even accuse their Democratic opponents of seeking to keep blacks in Welfare State bondage. But the actual data tells a very different story about the likely sources of Republican support.
The strength of this pattern may be seen at its extremes. Mississippi is the state with the highest black percentage and across all six elections its white population was the most likely to vote Republican, with the figures recently running at nearly the 90% level. Louisiana, Georgia, and South Carolina are generally clustered together as the next blackest in population, and in most elections their white populations were the next most likely to support the Republican ticket, although being sometimes exceeded by the whites of Alabama, the fifth or sixth blackest state during those decades.
By contrast, consider the three states with the largest non-white percentages: Hawaii, California, and New Mexico. The whites of the first two have actually been far less likely to vote Republican than whites nationwide, while those in New Mexico fall close to the national average. This tends to confirm the national statistical results that the widespread presence of non-whites, even in overwhelming numbers, seems to have little impact upon white voting behavior.
While I would not argue that black crime is the sole determining factor behind the racial polarization in white voting behavior, I do suspect it is one of the largest contributors. Empirically, the presence of blacks causes whites to vote the “law-and-order” Republican ticket, while the presence of Hispanics or Asians seems to have negligible political impact.
Nevertheless, we should remain cautious in interpreting these results. For example, although these national correlations are certainly substantial, they are almost entirely due to the weighting of the Southern states, in which blacks are almost 20% of the total population and racial tensions have traditionally been the strongest. In non-Southern states, the correlations are nil, perhaps partly because blacks are found in far smaller numbers, being less than 9% of the total.
Powerful quantitative evidence for social determinism may be dispiriting, and when the main determinant seems to be race, many Americans will choose to throw up their hands and ignore the statistical facts, simply hoping that these might somehow be proven incorrect. That is certainly their privilege, but for those individuals who prefer to grit their teeth and mine the data for contrary indications, there do exist a few interesting nuggets.
Weighted average correlations are a very useful summary statistic, but they neither tell the whole story nor do they preclude the existence of outlying cases, which might provide some insights on ameliorating the grim situation we have described. And it so happens that among our many dozens of major urban centers one of the most extreme race/crime outliers is neither small nor obscure: New York City. Our largest metropolis often has crime rates that deviate sharply from the usual urban pattern observed almost everywhere else.
Recall our earlier mention of the surprising absence of any correlation between urban population density and crime rates. Those summary statistics were correct, but they also hid some important variations and the null overall result was almost entirely due to the extremely high density and low crime rates in America’s largest city, combined with its huge population-weighting. If we excluded New York City from our calculations, the remainder of America’s major urban centers would demonstrate some moderately strong and fairly stable correlations between density and crime over the last dozen years; for example, density has generally had a positive correlation of around 0.35 with robbery rates.
Similar anomalies appear in the racial crime calculations that have been the central focus of our analysis. Based on its racial composition, we would expect New York City’s homicide rate to be some 70% higher than it actually is, with robbery and violent crime also being far more widespread. Cities like San Jose and San Diego may have homicide and violent crime rates only half that of New York City, but given the stark differences in their underlying demographics, it is New York City’s Finest who deserves praise for their remarkable effectiveness in crime prevention. Evaluating the apparent success or failure of urban law enforcement policies without candidly considering a city’s demographic challenges may lead to incorrect policy judgments.
Little of New York City’s success in crime prevention seems due to the relative size of its police force, which is roughly similar to those of Chicago, Philadelphia, Baltimore, and Boston on a per capita basis, and far below that of Washington, D.C., all cities whose crime rates reflect their demographics. So it appears that New York City’s crime-fighting methods rather than merely the number of its officers has been the crucial factor.
Ideas have consequences, as do attempts to avoid them. For most of the last twenty years, the policing methods implemented under mayors Rudolph Giuliani and Michael Bloomberg won enormous national praise as they so dramatically cut New York crime rates: murders dropped by over three-quarters. But during the last few years, some of these same policies have begun receiving widespread criticism among those pundits who may have forgotten just how bad things were two decades ago.
Our simple statistical analysis obviously does not allow us to disentangle the relative importance of the different factors behind New York City’s success. Since the early 1990s, the city implemented a “community policing” model as well as pioneering the rapid use of local crime data to pinpoint dangerous hotspots and allocate resources more accurately. But other elements of the package have included strict, even harsh policing methods, such as the widespread use of “stop-and-frisk” to reduce gun violence. Denouncing these techniques as unconstitutional or racially discriminatory may be perfectly justified, but those who do so must consider the trade-offs involved, including the very real possibility of a 70% rise in homicides if local policing effectiveness declined to levels found in the rest of the country.
Let us compare the demographic and crime trends of New York City and Washington, twin abodes of our East Coast urban elite. Between 1985 and 2011, Washington’s homicide rate dropped by 26%, robbery fell 27%, and violent crime in general was cut by 30%; but the city’s black population also dropped by 27% during this same period. Meanwhile, New York City’s corresponding declines in crime were far greater, 67%, 78%, and 67% respectively, but were accompanied by only a small 7% decline in black numbers. For all these serious crime rates to decline at nearly ten times the rate of their primary racial determinant is absolutely remarkable, a combination that left the city an exceptional outlier among America’s major urban centers.
Put another way, if America’s other cities with large black populations had somehow managed to achieve the same surprisingly low crime rates as New York City then most of the high racial crime correlations that have been the central findings of this article would disappear. Conversely, if New York City were excluded from our current national statistics, many of the existing racial crime correlations would exceed 0.90. These are objective facts and well-intentioned analysts who sharply criticize New York City policing methods should recognize that they may face some unpalatable choices.
Perhaps further research would establish that the widely-lauded elements of local police practice are the ones primarily responsible for such results, and the more controversial methods may safely be eliminated without negative consequences. But for whatever combination of reasons, the overall results achieved by New York City have been quite remarkable and caution should be exercised before drastic changes are made in such a successful model.
Obviously New York City is not the sole positive outlier on these crime statistics, though it is by far the most significant, both because of its size and the magnitude of its deviation from the predicted results. If we examine the 2011 homicide rates for our set of sixty-six large cities, seventeen of these were at least 30% below the projected trendline, with four cities—Charlotte, Raleigh, St. Paul, and Virginia Beach—achieving even better results than New York City. But many of these successful cities have numerically small black populations, and the total for all seventeen combined is not much larger that of New York City alone. One intriguing fact is that although fewer than one-third of the all our large cities lie in the South, these Southern cities account for over two-thirds of those particularly successful examples, and a roughly similar pattern applies both for other crime rates and for other recent years. The exact mix of cultural, socio-economic, or demographic factors responsible for such notable Southern success in achieving relatively low urban crime rates is unclear, but might warrant further investigation.
Over the last decade or two, liberal intellectuals have regularly denounced their conservative opponents for allowing ideological considerations to trump objective facts, sometimes styling themselves the “Reality-Based Community” as an ironic riposte to the foolish criticism of a top Bush Administration official. Many of these liberal accusations have considerable merit. But individuals who claim to accept reality undercut their credibility if they pick and choose which portions of reality they acknowledge and which portions they carefully ignore. Our academic and media elites should not avoid factual evidence that they dislike.
Consider that over one-quarter of all the urban black males in America have vanished from our society, a loss-ratio approaching that experienced by Europeans during the Black Death of the Middle Ages. Yet these astonishing statistics have largely remained unreported by our major media and hence unrecognized by the general American public. Should the medieval scribes of the Fourteenth Century have ignored the annihilating impact of the bubonic plague all around them and merely confined their writings to more pleasant news?
It is said that very young children sometimes believe they can hide themselves by covering their eyes, and that seems to be the general approach taken by our major media to the unpleasantly grim racial crime statistics analyzed in this article. But the reality continues to exist whether or not we ignore it.